Core Concepts
This survey provides a comprehensive introduction to the rapidly developing field of risk measures and their applications in diverse areas, including engineering design, data-driven problems, and decision making under uncertainty. It highlights the central role of superquantiles (conditional value-at-risk) in unifying various threads and connecting concepts of risk, regret, deviation, and error.
Abstract
The survey begins by illustrating how problems of decision making under uncertainty arise in diverse areas, such as engineering design, control, statistics, and machine learning. It defines measures of risk and provides concrete examples, making connections with utility theory.
The core of the survey focuses on superquantiles (also known as conditional value-at-risk, average value-at-risk, tail value-at-risk, and expected shortfall). It describes the historical development of superquantiles, their central role in many areas of operations research, engineering, and statistics, and provides a behind-the-scenes description of the derivations that led to an influential formula for superquantiles.
The survey also discusses a duality theory and connections with distributionally robust optimization. It introduces the terminology "measure of reliability" for failure probabilities, buffered failure probabilities, buffered probabilities of exceedance, and related concepts. Throughout, the survey focuses on computational aspects and implementable algorithms.
The survey concludes with extensions and open questions, including measures of reliability, dynamic and multi-stage optimization, and other challenges and open problems in the field of risk-adaptive stochastic optimization.
Stats
"Uncertainty is prevalent in engineering design, data-driven problems, and decision making broadly."
"Real-world data sets tend to be noisy, biased, corrupted, or simply insufficiently large."
"The concept of risk measures provides a mathematical framework for unifying and understanding the various threads and how they connect."
"Risk measures furnish a rich area for nonlinear analysis as well as opportunities for efficient computations."
Quotes
"The vast literature on risk measures developed over the last 25 years is a testimony to the potency of the framework, both theoretically and practically."
"Risk measures capture all these possibilities and many more. Through choices of probability distributions, risk measures allow us to incorporate data and other information about the possible values of ξ and their likelihoods."
"Supported by convex analysis, risk measures furnish a rich area for nonlinear analysis as well as opportunities for efficient computations."